The Research Of Content-based Image Retrieval Technique

Content-Based Image Retrieval (CBIR) is a kind of technique for retrieving images on the basis of automatically extracting visual features such as color, texture, and shape etc. On the technique condition of development of network, database and multimedia etc., it has been an active research area in recent years.Chapter 1 is the exordium. In the chapter, we give an introduction to the Image Retrieval, including its background, application and some representative CBIR system.Chapter 2 expounds the detail of the important branch of Information Retrieval—Content-Based Image Retrieval. It includes some expression of visual features about an image, the similarity between two images and relevance feedback technique.In Chapter 3, we present a new image retrieval method, which is based on color distribution. To express accurately the information of color distributing to a image, in this part, two new color feature of color-based image retrieval method are proposed, which both dominant-color matrix and dominant-color pair matrix. After first searching image with the two features, the find effect is more improved by two new different relevance feedback techniques that have something to do with the two color features respectively.In Chapter 4, a new algorithm, which is based on Hide Markov Chain (HMM) model, was presented to extract texture and shape feature of an image. To attain the spatial information, the gray of the image in HSV mode are firstly divided into 4×4 blocks which are classified into 3 status, flatness, texture and edge status. Then each gray image is translated into a matrix composed of those 3 status values. After that the status matrix is translated into dimension status sequence, the transition probability matrix of the sequence is calculated as the image's spatial distribution information. In order to match similarity between two images, the coefficient is used to measure the distance of different transition probability matrixes.Chapter 5 introduced the image retrieval method based on multi-feature. We use two eigenvectors both dominant-color matrix and transition probability matrix as whole target-image's feature to retrieval the images.Chapter 6 explains the structure of my CBIR system, which is used to test the algorithm's validity in the chapter 3 and 4, and illustrate the modules' function.The last chapter first gives a summary to the thesis, then presents some problems on the CBIR technique, which remain to be resolved, and some possible research directions in the future are pointed out.